EduSathi: AI-Powered Handwritten Answer Sheet Grading Platform

Evaluation workflow for Academic Institutions & Universities. Apply for pilot
0+
Registered Users
0+
Exams Created
0+
Answer Sheets Graded

See a Real Grading Run

Question Paper
Question
Teaching Material
Teaching material

“I drew the histogram correctly. Where did I lose the half mark?”

Student answer
Student asks

All six bars, both axes, and the scale are present. Why is the score 3.5 out of 4?

EduSathi report
Question score3.5 / 4
X axis label with unit0.5 / 0.5
Y axis label with unit0.5 / 0.5
Correct and uniform scale1 / 1
Accurate plotting of all bars1.5 / 1.5
Graph title was missing0 / 0.5

The bars were correct. The half mark was deducted only because the required graph title was not written.

Student challenges the mark: “Why did these two answers receive 0 out of 2 and 0.5 out of 2?”

Question 2(d) · Extended Binary Tree · 0 / 2

The answer described a balanced binary tree. It did not state that null links become external nodes or that every internal node has exactly two children.

Question 2(e) · Depth First Search · 0.5 / 2

The answer mentioned depth wise exploration and included a diagram, but omitted stack or recursion and visited tracking or cycle detection.

  1. How do I verify every page total?
  2. How do I avoid carry forward errors?
  3. How do I finish checking within the deadline?
  4. How do I answer every marks complaint?

Exam results take weeks or months.

What EduSathi provides

Teacher controlled AI grading

Teacher textbook materialTeacher material

Teacher's textbook and material

Question paper, textbook, model answer, marking instructions.

3.5 marks

Everything editable. Teacher final authority.

Model answer, step-wise marks, reasoning, and every student mark.

Traditional2–3 weeks
EduSathiAvg. 3 days*

From 2–3 weeks to an average of 3 days*

Significantly reduces turnaround time across scanning, evaluation, moderation, and report generation.

*Turnaround varies by batch size and teacher review requirements.
Grading capabilities
Step 11 / 1
Step 20.5 / 1
Step 31 / 1

Step-wise marks

Marks are given step by step, not as an unexplained final total.

मराठीहिन्दीதமிழ்తెలుగుಕನ್ನಡMore…

Indian regional languages

Supports handwritten answers in 10+ Indian languages.

StrictModerateLenient

Teacher-set grading style

Teachers can test a few papers before the full batch and choose Strict, Moderate, or Lenient grading.

Diagram grading

Evaluates diagrams as well as written answers.

The Full Workflow

From creating your exam to downloading graded reports in seven steps.

1
Create Account
Sign up with your email
→↓
2
Create Exam
Set subject, class, total marks
→↓
3
Submit Materials
Upload question paper, marking scheme & textbook
Question Paper
Question paper
Textbook Context
Textbook page
4
AI Drafts Model Answer
System creates marking scheme from textbook
→↓
5
Teacher Reviews & Approves
Edit, adjust, finalise marking scheme
6
Upload Student Sheets
Upload scanned/photo answer sheets (ZIP or PDF)
→↓
7
Download PDF Reports
Student wise reports after teacher approval
Student Answer Sheet
Student answer sheet
Sample PDF Report | Real Output
PDF grading report↓ Download Sample Report PDF
AI and human examiner score comparison for the Fluid Mechanics pilot
IIT Madras pilot accuracy

Fluid Mechanics grading benchmark

AI evaluation compared with human examiner scores across 9 anonymised handwritten answer sheets.

0.92Correlation
4.8%Average deviation
−0.11Mean bias
View accuracy report →
OCR case study

EduSathi AI pipeline beats frontier models in faint-handwriting OCR

EduSathi's preprocessing recovered a more complete and accurate transcription from a faint, low-contrast handwritten answer sheet.

Read more →
Built with Google Gemini

Gemini Vision pipeline vs a model wrapper

The model is Gemini on Vertex AI. The edge is the exam pipeline — native-page vision, teacher model-answer, and batch context cache — which is why this tracks a human examiner and beats raw frontier vision on faint Indian handwriting.

EduSathi · Gemini Vision on VertexRaw frontier visionChatbot + OCR wrapper
InputNative PDF / page images to Gemini; Drive file_uri so Gemini reads from Google storageImage dump, no exam preprocessingOCR text pasted into a chat model
Faint handwritingPreprocess + Gemini Vision recovered more complete text than the frontier models in our case studyDrops low-contrast strokesOCR fails first, model never sees the ink
MarksStep marks constrained by the teacher model-answer; teacher can edit every cellOpen-ended score, no schemeRubric in the prompt only
Human examiner (IIT Madras AS1020)0.92 correlation · 4.8% average deviation · −0.11 mean biasNot our published protocolNot our published protocol
Google CloudVertex Flex, context cache, Cloud Run / Cloud Run functions (Lambda-class workers)Vendor API onlyVendor API only

Accuracy figures are from the published Fluid Mechanics and faint-handwriting case studies — not a lab leaderboard.

First page of the EduSathi brochure
EduSathi brochure

EduSathi's AI grading workflow for academic institutions.

Custom pricingLet's talk about your requirements.View pricing →

Meet Our Team

Team →

Pre-incubated at IIT Madras. Built with Google Gemini on Google Cloud.

Nirmaan
NIRMAAN
The Pre-Incubator, IIT Madras
Google Cloud
Google Cloud
Vertex AI · Cloud Run
Gemini
Google Gemini
Built with Google Gemini
DeepMind
Google DeepMind
Gemini models
Google for Startups
Google for Startups
Startup Buildathon 2026
Y

Special thanks to Y Combinator for infrastructure credits and resources powering EduSathi.

Frequently Asked Questions

Common questions about EduSathi.

FAQ